Machine learning in trading: theory, models, practice and algo-trading - page 505
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
The Nobel Prize in Physics was awarded yesterday. I remember that you have Einstein as a loon, but look - did you measure normally? You never know...))
https://indicator.ru/article/2017/10/03/za-chto-dali-nobelevskuyu-premiyu-po-fizike-2017/
Don't confuse my words with your generalizations. If you want to remind me of my words, quote me, but if you drew your conclusions from my words, just say so: "I drew conclusions from your words that you think that Einstein is a horse" and it will not be my words but your conclusions based on which it is not clear. Einstein is a brilliant figure in world science, he should manage to write the theory of relativity for some and the special theory of relativity for others.
With respect.
What's the lib, where did the wood come from? (trees)
Sincerely.
Thank you for your reply! Very informative for me.
I take it bias is a displacement neuron? The description says that it helps when the input is zero. What do you think the bias neuron is used for, and what kind of weights should it take? Basically it's just a weight.
Is it better to check the threshold value after sigmoid transformation or after?
> It says in the description that it helps when there is zero entry
It has nothing to do specifically with zero input; the bias gives the neurons more accuracy. It has to be (c).
When training neuronics - offset is optimized at the same time as other weights.
> How is it better to check the threshold value after sigmoid transformation or after?
If an activation function is used for output neurons, then check values on different outputs or check values with thresholds after the activation function (after the sigmoid).
I should add that in hidden neuron layers it is mandatory to use some kind of activation function, while activation is not so mandatory for output neurons (the last layer), for example I don't use activation in regression (i.e. linear activation).
I meant that for example in mql4 it was possible to optimize simultaneously up to 15 parameters, and in mql5 more.
It appears that one layer is adjusted, then the second one with optimized first layer, etc. It would be nice if we could optimize all the layers at once, but the computing power is not enough.
I have an assumption that when layers are optimized one by one, in this case some pattern is no longer seen by the system.
Even if one layer is parsed, the next layers will be based on the assumptions of the first layer.
It is possible to train a neuron by layers, it is done when training neurons with a very large number of layers. But this is a last resort... if there are only a couple of layers, it is better to optimize all weights at once.
Looking for neuronka weights with genetic optimizer in MT is not a good idea. Neuronka will just be fitted to existing examples, and will not correctly predict new data. Read about neuronka training using crossvalidation, it is a must to know and do. The way to train them will help them to recognize pictures and letters, but for forex this is not enough, you need to invent your own ways, like how to test expert advisors with a Walk-Forward test.
I want to ask distinguished participants on this subject
The question, as I see it, is key: how much does the real price data differ from the SB? If I understand correctly, the greater the difference, the more opportunities to squeeze out a profit. And vice versa, up to "no difference - no profit".
They differ from random just a little bit...
If i look at the chart, i see an example of profits in eurusd m5, model + advisor at the beginning of each bar predicts a long or short and either keeps the same deal or reverses it.
The first 10 000 bars are the data on which the model was trained, then 10 000 new bars for the model.
Profit is shown as the sum of points, for example profit 0.08 is = 0.08000 = 8000 five-digit points.
Three charts - without spread, and with a small spread (2 and 4 five digits). Commissions and swaps I do not take into account at all, there is not so clear trade simulation.
The gurus here in the thread squeeze much more accuracy out of my data, trade probably can still be improved if you know how.
Which is interpreted as - manipulator, cheater. God be with him))))
Are the measurements okay? + Attitude towards the study of the laureates...
Sincerely.
They differ from random just a little bit...
Here is a sample profit chart when trading on eurusd
I don't understand what you want to show. Are they different or not? And if they are different, in what way?
Maybe, I have understood your question wrong, correct me. You're saying that if you have managed to build a profitable TS, then it is different. I totally agree with that if. But it doesn't follow at all from the charts that you have succeeded. No one knows if anyone has succeeded or not. That's why I was asking about the differences between SB and CBR that the MoD can find.
That is, the approach is not the reverse: there is a profit, so it is different. Because we do not know whether the profit is legitimate. But the direct approach. When IRM reveals some regularities, but it is not certain that they can be used to get a profit. However, it somewhat increases the probability to achieve it at some point.
fxsaber:
You got it right.
I agree, the reasoning is logical, just because somewhere on the Internet the profit on the chart goes up does not mean that forex is not accidental.
https://github.com/RandomKori/ForexTF I switched to the Tensorflow library. There is more documentation for it. It can be connected to MT. I compiled 64 bit library for Windows. I put it up. Don't let the extension of so library bother you. It works fine on the Windows and is used in Visual Studio. I am just spoiling my own implementation of ResNet for forex and forts. The results are encouraging. I am still struggling with retraining. My immediate plans are to write an indicator for ResNet. All code is on github (as always).
Let's have a serious talk. I need your professional advice. Who do you think is going to take whom - Odin or Thor?????
Hahaha ha)) Dimitri's a real piece of work.)
I support you in the argument, it's all a setup
Hi all!!! Guys, tell me how to delay the calculation of the indicator. A new bar opened and the indicator needs to be calculated after 30 seconds!!!!!
Or tell me where it is written or an example, I searched all over, I can not find :-(